Predicting unbound phenytoin concentrations in patients receiving valproic acid: a comparison of two prediction methods

OBJECTIVE: To compare the predictive performance of 2 equations that estimate unbound (free) phenytoin plasma concentrations when valproic acid (VPA) and phenytoin are administered concurrently. DESIGN: Eighty-eight adults receiving VPA and phenytoin concurrently were included in the study. Steady-state plasma concentration measurements of total phenytoin, total VPA, and unbound phenytoin were collected prospectively in the inpatient group (group 1) and retrospectively in the outpatient group (group 2). Using the equations developed by Haidukewych and May, unbound phenytoin concentrations were calculated. The mean predicted unbound phenytoin concentrations then were compared with mean actual unbound phenytoin concentrations measured in the laboratory. Identical assays were performed to measure unbound phenytoin, total phenytoin, and VPA from each patient group. SETTING/PARTICIPANTS: Antiepileptic drug concentration measurements were collected from 43 inpatients (mean age 34.8 y) from the epilepsy unit at Abbott Northwestern Hospital and 45 outpatients (mean age 37.3 y) at the MINCEP Epilepsy Care clinic, Minneapolis, MN. MAIN OUTCOME MEASURES: Mean prediction error (MPE) and mean squared error (MSE) were calculated to determine which equation was the least biased and most precise in predicting unbound phenytoin when total VPA and phenytoin concentrations are known. Linear regression of predicted unbound phenytoin on measured unbound phenytoin values determined the correlation coefficients (r). A paired Student's t-test also was performed comparing mean predicted unbound phenytoin concentration with mean actual unbound phenytoin concentrations in both groups. RESULTS: The MPE from May's equation was -0.49 and -0.45 for groups 1 and 2, respectively; using the Haidukewych equation, MPE was -0.02 and 0.08 for groups 1 and 2, respectively. The MSE using May's equation was 0.34 for both groups. Using the Haidukewych equation, group 1 MSE was 0.07, and for group 2, 0.12. Correlation coefficients were more than 0.91 (p < 0.001) from each equation in both patient groups. In group 1, mean actual unbound phenytoin concentration was 2.02 micrograms/mL; May's equation predicted 1.52 micrograms/mL (p < 0.001) and the Haidukewych equation predicted 2.00 micrograms/mL (p = 0.64). In group 2, mean actual unbound phenytoin concentration was 2.10 micrograms/mL; May's equation predicted 1.65 micrograms/mL (p < 0.001) and the Haidukewych equation predicted 2.18 micrograms/mL (p = 0.11). CONCLUSIONS: Haidukewych's equation predicts unbound phenytoin concentrations with the least bias and most precision with statistical significance. May's equation consistently underpredicted unbound phenytoin concentrations. Because unbound phenytoin fraction is not constant (and usually more than the expected 10%) in patients comedicated with VPA, unbound phenytoin concentrations cannot be predicted even though total VPA and phenytoin concentrations are known. If unbound phenytoin concentrations cannot be readily measured, Haidukewych's equation is a reliable predictor of unbound phenytoin concentrations.